Brian Zhou¹ and Alvan Caleb Arulandu², ¹USA, ²Harvard University, USA

The United Nations (UN) is the foremost international body helping uphold world peace through peacekeeping missions, ranging from deployments that enforce peace treaties, monitor conflicts, and protect civilians; However, determining when and how to intervene is complex. The updated UN General Debate Corpus (UNGDC), cataloging every speech from the UN's inception in 1946 to 2022, is a treasure trove of national policy, as the UNGD is the only body where every country can speak. We propose a discourse-driven intervention recommendation framework that categorizes ongoing conflicts based on UN precedent and recommends the magnitude of funds and forces that should be committed to addressing a conflict. We employ natural language processing techniques to tokenize, preprocess, and analyze word stem frequencies in the UNGDC, generating a timeseries of the number of UN mentions for any given country. Paired with historical analysis, we show that debate in the UNGDC is a potent indicator to determine UN intervention and response mechanisms for conflicts in Africa; further, by aggregating mention statistics across periods of active conflict, we provide quantitative backing for the correlation of mention dynamics and the presence of an active conflict, for a given country. Finally, we present and test an interpretable, shallow decision tree model that can perform intervention type classification and response magnitude recommendation with 91.7% accuracy. Our results, established by computational experiments and statistical testing, suggest that corpus analysis and broader computational diplomacy methods can drive intervention recommendations to improve the UN's decision-making.

United Nations, conflict resolution, natural language processing, decision trees, computational diplomacy, UN peacekeeping, UN General Assembly General Debate, UN General Debate Corpus, political communication